The Artificial Inventor: Amending Intellectual Property in the Age of AI

The rapid growth of artificial intelligence (AI) technology has garnered increasing public attention following the release of OpenAI’s ChatGPT. Released in late 2022, ChatGPT is an artificial intelligence chatbot that can write essays, solve complex math equations, and debug code. The abilities of ChatGPT, however, are not limited to these straightforward tasks but also include far more complex and subjective functions. Tech monolith Microsoft, one of OpenAI’s largest investors, claims that developers can use ChatGPT to, “[enhance] existing bots to handle unexpected questions, [recap] call center conversations to enable faster customer support resolutions, [create] new ad copy with personalized offers, [automate] claims processing, and more.”[1] And this is in addition to other applications such as data analysis in healthcare, accounting for small businesses, and lending for banks [2]. ChatGPT is one of the world’s most exciting technologies, and it offers a wealth of different possibilities for individual consumers, firms, and governments. What makes ChatGPT especially exciting is that it is a type of generative AI, a rapidly growing class of AI technology that uses large quantities of past data to generate original content. ChatGPT has marked the beginning of a new chapter in artificial intelligence and technological development as a whole, and many more products and improvements will surely follow.

Despite its impressive accuracy and speed, AI remains an experimental tool with a variety of significant risks. There are three primary points of concern: bias, privacy, and intellectual property (IP). First, many argue that because AI is trained on data generated by humans, it will necessarily share the same biases and prejudices that exist in human society. As Harvard Professor Michael Sandel argues, “We are discovering that many of the algorithms that decide who should get parole, for example, or who should be presented with employment opportunities or housing…replicate and embed the biases that already exist in our society.”[3] AI, at least for the time being, contains many of the same biases that humans have, so consumers of AI cannot view it as an infallible scientific solution to the widespread issue of human bias. Second, because AI requires a vast bank of data to train on, there is growing scrutiny regarding what data AI can and should be able to access, raising a series of complex concerns about privacy. For example, some argue that AI-producing firms abiding by the European General Data Protection Regulation (GDPR) — which applies to any processing of information that can identify individuals — must only process data if they have a lawful basis for doing so and must undergo data protection impact assessments (DPIAs) to ensure risk compliance [4]. Finally, while there has been much discussion about AI bias and data privacy, relatively little has been written about AI’s relationship to intellectual property. For a field driven by innovation and original ideas, one would think that ownership of these powerful new technologies would be a primary talking point. Only recently, though, did the U.S. Copyright Office begin paying serious attention to AI technologies, releasing updated guidelines and launching a program to learn more about the public view on AI technologies and copyright law [5]. For the vast amount of power and influence that AI holds, this indifference is concerning.

The growth of AI technology must be accompanied by appropriate IP legislation in order to set clear expectations and boundaries for future entrepreneurs and innovators. For the first time in the history of intellectual property law, humans have developed a technology that can formulate its own works and products. As a result, traditional intellectual property law — which only designates humans as inventors — must be updated to reflect this change [6]. Instead of instituting a blanket ban on AI as inventors, U.S. IP law should instead aim to distribute credit fairly. In a practical sense, this would mean giving credit to AI systems, which could potentially surpass human intelligence in the future, while also providing credit to the owner(s) of the AI system and the data on which the AI trained. By distributing intellectual property in this way, the law attributes some credit to the machine that actually developed an invention and avoids setting the precedent of assigning all subsequent inventions of an AI to its creator or owner, who likely contributed little to developing these inventions themselves.

Some legal scholars have objected to allocating credit to the owners or inventors of AI systems, claiming that “humans operating inventing software are not inventors.”[7] However, they refer back to traditional IP law in order to justify their arguments, a practice that is at odds with the groundbreaking nature of AI technology. Distributing intellectual property is not a radical idea. Some scholars have already suggested that patents could be awarded to “the AI developer, the person directing the AI and the owner of the data used to train it.”[8] On the other hand, distributing intellectual property to AI systems themselves is indeed a radical idea. Thus, in order to support the exciting growth of AI technology while also maintaining the foundations of U.S. IP law, future U.S. IP law should define a special category for AI technology and divide the intellectual property rights among the AI itself, the creator/overseer of the AI, and the owner(s) of the data on which the AI trained.

U.S. IP law has fallen behind the rapid growth of AI technology. For example, consider the inconsistent and insufficient treatment of computer scientist Stephen Thaler’s patent applications. Thaler submitted two patent applications in 2019 that designated a self-developed and patented AI system — Device for the Autonomous Bootstrapping of Unified Science (DABUS) — as the only inventor [9]. However, the U.S. Patent and Trademark Office (USPTO) denied his application on the grounds that it lacked a valid inventor [9]. In Thaler v. Hirshfeld (2021), the District Court for the Eastern District of Virginia also rejected Thaler’s stance that an AI system was a valid inventor [10]. On appeal, the U.S. Court of Appeals for the Federal Circuit confirmed their reasoning in Thaler v. Vidal (2021), “we, too, conclude that the Patent Act requires an ‘inventor’ to be a natural person and, therefore, affirm.”[12] On the international level, Thaler has also faced rejection from the United Kingdom, European Patent Office, and Australia, though South Africa did approve one of Thaler’s patent applications. As such, the international interpretation has been inconsistent. In addition, the current approach that rejects any patent applications from AI systems is insufficient. It creates a gray area for AI-created inventions, forcing inventors to choose whether to erroneously attribute credit to themselves or risk leaving inventions unpatented. This uncertainty only serves to weaken innovation and stifle the development of new technologies. Thaler has now filed a petition asking for the Supreme Court to review his case [13].

Such a radical change in IP law is not unprecedented. Indeed, IP law has adjusted significantly in the past to account for technological developments. In Diamond v. Chakrabarty (1980), the Supreme Court ruled that the scientist Ananda Chakrabarty could patent his discovery of an artificially-created bacterium used to break down components of crude oil [14]. In the majority opinion, Justice Burger wrote that “[Chakrabarty’s] micro-organism plainly qualifies as a patentable subject matter. His claim is not to a hitherto unknown natural phenomenon, but to a non-naturally occurring manufacture or composition of matter – a product of human ingenuity.”[15] While the patent eligibility for certain inventions has caught up with modern innovation, the patent eligibility for certain inventors still lags far behind. By deciding Diamond, the Supreme Court defined clear regulatory expectations for the biotechnology and pharmaceutical industries. Perhaps it can do the same for AI technology.

The general structure of U.S. patent law, an important subset of U.S. IP law, is centered around inventions, inventors, and process [16]:

• “The term ‘invention’ means invention or discovery.”[17]
• “The term ‘inventor’ means the individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention.”[18]
• “The term ‘process’ means process, art or method, and includes a new use of a known process, machine, manufacture, composition of matter, or material.”[19]

Combining these three definitions, the U.S. Code states that “whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.”[20] These definitions are still fundamentally sound for most applications. However, these definitions, especially the concept of “inventor,” do not apply as readily to AI technology and innovation. For example, suppose an AI technology such as ChatGPT has formulated a revolutionary cancer drug that offers a remarkable cure for all cancer patients. The intellectual property rights for this invention are clearly very important. What is the invention? This is simple: a new cancer drug. Who is the inventor? This is less clear. Is it the AI system itself or the developer behind the system? Is it the data on which it was trained? Finally, and most unclearly, what was the process? Was the training of the AI model the process? Or was it the AI’s own “creation” of the drug? If so, would the AI not be the inventor? In trying to fit AI inventions into the existing frame of IP law, a seemingly endless number of thorny questions arise.

U.S. IP law, therefore, should designate a special category for intellectual property cases regarding AI technology to more accurately reflect its specific nature. These kinds of custom-built, sui generis laws are often used to address works not covered by the main four IP doctrines: copyright, industrial designs, trademarks, and patents [21]. In doing so, U.S. IP law would maintain its fundamental definitions while also acknowledging the extraordinary nature of AI technology. Within this category for AI technology, U.S. IP law should distribute intellectual property rights among three distinct parties: the AI inventor, the creator or owner of the AI system, and the owner(s) of the data with which the AI was trained. To many, the very idea of giving a non-human entity credit for an invention may seem strange, dangerous, and pointless. However, it is a necessary step in order to acknowledge the original discovery by the AI system and to avoid giving undue credit to the original human AI inventors. Consider the scenario in which creators or owners of AI systems were granted the complete intellectual property rights to any and all of their AI’s innovations. Then, scientists and engineers would focus primarily on developing AI systems instead of focusing on their respective fields and industries in order to reap the benefits of AI’s incredible power and efficiency. While this shift may seem optimal to some, there is a fundamental underlying issue: training data. AI systems are largely trained on human-generated data, so these changing incentives may distort and undermine future AI innovations by corrupting the very data on which AI systems are trained. However, this is not to say that AI systems should receive all the intellectual property rights. The creator of the AI system should also receive some credit. In this way, the law properly gives credit to the developers of this unique technology and incentivizes scientists and engineers to pursue AI development with the knowledge that they will share in the rewards of the AI’s subsequent inventions. Finally, IP law should assign intellectual property rights to the owner(s) of the AI training data. This is an often overlooked aspect of AI technology but nonetheless one of the most vital. Without training data, AI systems would be unusable. While some may argue that IP law does not attribute intellectual property rights to all of an invention’s sources and influences (of which there are surely many), AI is a clear exception. An AI’s training data is a core aspect of its function and ability, and as such, the owner(s) of that data should be compensated accordingly.

By splitting the intellectual property into three parts, the law would avoid a perilous loophole. When attributing intellectual property rights solely to AI systems, IP law allows AI creators to simply transfer the AI’s rights to themselves. With this loophole open, AI developers would be able to apply for patent applications designating their AI systems as inventors but subsequently transfer those rights to themselves, thus rendering the AI-inventor designation as a mere symbolic gesture. Indeed, Thaler has already attempted to exploit this loophole with his DABUS patent applications [22]. By directly distributing intellectual property rights to the creators themselves and introducing another party, the owner(s) of the AI training data, IP law would essentially close this loophole. A natural follow-up question to ask would be regarding who controls the AI’s intellectual property rights. While this three-pronged approach may appear clear and straightforward, there is a fundamental change that must also happen for this approach to work. Since AI technology is not conscious, it would be difficult to argue that AI should maintain full control of its intellectual property rights just as a human would. Thus, an ideal solution would be to auction off the AI-held intellectual property to any willing buyer except for the creator or owner of the AI system itself, with the proceeds to be reinvested in the development of the AI system. This strategy would transfer the AI’s intellectual property to a human who could maintain full, conscious control while still distinguishing its component from that of its creator or owner. In this way, the intellectual property rights would still be shared among the three parties and the owners of the AI technology would not be able to usurp the AI’s intellectual property rights. Indeed, this proposal is in its early stages, and there are surely many details to be analyzed and discussed. Furthermore, the specific distribution of the intellectual property rights remains a topic for discussion. Nevertheless, the most important requirement is that each of the three entities discussed receive a portion of the intellectual property and its subsequent benefits.

Many do not realize the vital role IP law plays in driving entrepreneurship and innovation. It creates incentives, sets boundaries, and shapes entire industries. As such, revising IP law to better incorporate AI technology and its innovations should be a legislative priority. With the rise of technology such as ChatGPT, the public is just beginning to see the true potential of generative AI, and AI research, development, and investment are likely to grow rapidly in the coming years. This rise in AI development will certainly be accompanied by a similar increase in AI-related intellectual property cases, so up-to-date IP laws will be vital for both managing and stimulating this thriving field. By designating a special category in IP law for AI technologies and distributing intellectual property rights among the three aforementioned entities, IP law can acknowledge the extraordinary nature of AI, bolster incentives for innovation, and maintain its traditional core definitions. Given the long-standing issue of gridlock and political polarization in the U.S. Congress, writing and passing effective legislation on a fast-changing field like AI may seem unfeasible. And with the Supreme Court’s lengthy docket of important cases, some may argue that AI-related IP law should wait. Nevertheless, despite the potential legislative and judicial challenges, legislators, legal scholars, and AI developers alike must work together to craft clear and fair IP laws for AI. With the immense potential of AI technology, the effort will be well worth it.

Bibliography

1. Boyd, Eric. “ChatGPT Is Now Available in Azure OpenAI Service.” Microsoft.com, https://azure.microsoft.com/en-us/blog/chatgpt-is-now-available-in-azure-openai-service/. Accessed April 14, 2023.

2. Pazzanese, Christina. “Ethical Concerns Mount as AI Takes Bigger Decision-Making Role.” Harvard Gazette, October 26, 2020.

3. Ibid.

4. Massey, Rohan, and Clare Sellars. “The GDPR and AI: Ensuring Data Protection From the Start.” Bloomberg Law, October 22, 2020. Accessed April 14, 2023, https://news.bloomberglaw.com/privacy-and-data-security/the-gdpr-and-ai-ensuring-data-protection-from-the-start-16.

5. “NewsNet Issue 1004 | U.S. Copyright Office.” Accessed April 14, 2023.

6. 35 U.S.C. § 115.

7. Schuster, Michael. “Artificial Intelligence and Patent Ownership.” Washington and Lee Law Review 75, no. 4 (February 19, 2019).

8. George, Alexandra, and Toby Walsh. “Artificial Intelligence Is Breaking Patent Law.” Nature 605 (May 26, 2022).

9. Akin Gump Strauss Hauer & Feld LLP.. “Federal Circuit Confirms ‘Inventor’ Must Be Human, Not AI.” Akin, August 11, 2022, https://www.akingump.com/en/insights/alerts/federal-circuit-confirms-inventor-must-be-human-not-ai. Accessed April 21, 2023.

10. Ibid.

11. Thaler v. Vidal, No. 21-2347 (Fed. Cir. 2022).

12. “Federal Circuit Confirms ‘Invenstor’ Must be Human, Not AI.”

13. “Can AI Invent? High Court Asked to Take Up Thorny Patent Issue.” Accessed April 21, 2023.

14. Diamond v. Chakrabarty, 447 U.S. 303 (1980).

15. Ibid.

16. 35 U.S.C. § 100.

17. 35 U.S.C. § 100(a).

18. 35 U.S.C. § 100(f).

19. 35 U.S.C. § 100(b).

20. 35 U.S.C. § 101.

21. George and Walsh, “Artificial Intelligence is Breaking Patent Law.” Nature.com, May 24, 2022, https://www.nature.com/articles/d41586-022-01391-x. Accessed 14 April, 2023.

22. “Federal Circuit Confirms ‘Invenstor’ Must be Human, Not AI.”